hmlet's Introduction

Paper Accepted to WSDM'22

Title: Linear, or Non-Linear, That is the Question!

How to Use

Docker environments

docker pull pytorch/pytorch

Docker run

docker run --gpus all -it --rm --privileged -v {local_path}:/HMLET pytorch/pytorch bash -c "pip install pandas && pip install scipy && pip install sklearn && pip install tensorboardX && pip install openpyxl && cd /HMLET && {train_model_command}"

Train model

python train.py --dataset {dataset_name} --model {model_variants}

Methods Proposal Background and Purpose

Which embedding propagation (linear & non-linear) is more appropriate to recommender systems?

Methods

HMLET (Hybrid-Method-of-Linear-and-non-linEar-collaborative-filTering-method)
- Dynamically selecting the best propagation method for each node in a layer using gating networks.
Four variants of HMLET: HMLET (End), HMLET (Middle), HMLET (Front), HMLET (All)
- Four variants of HMLET in terms of the location of the non-linear propagation.
HMLET (End)
- HMLET (End) shows best performance among these variants
- Focusing on gating in the third and fourth layers
- The detailed workflow of HMLET (End)

hmlet's People

Stargazers

Watchers

hmlet's Issues

I want to ask you about the idea of the article, that is, why you should use Gumbel-softmax - if I understand correctly, the input in this article is not sampling.Even with softmax, there should be no differentiable problems without reparameterization.

From the perspective of solving argmax‘s problem, I think softmax with temperature τ should also achieve a similar effect.

My understanding is that according to the statement in Alg1 and 2, a one-hot control variable needs to be output, but the specific explanation is still vague.

Best wishes！

Recommend Projects

qbxlvnf11 / hmlet Goto Github PK

hmlet's Introduction

Paper Accepted to WSDM'22

Title: Linear, or Non-Linear, That is the Question!

How to Use

Docker environments

Docker run

Train model

Methods Proposal Background and Purpose

Which embedding propagation (linear & non-linear) is more appropriate to recommender systems?

Methods

HMLET (Hybrid-Method-of-Linear-and-non-linEar-collaborative-filTering-method)

Dynamically selecting the best propagation method for each node in a layer using gating networks.

Four variants of HMLET: HMLET (End), HMLET (Middle), HMLET (Front), HMLET (All)

Four variants of HMLET in terms of the location of the non-linear propagation.

HMLET (End)

HMLET (End) shows best performance among these variants

Focusing on gating in the third and fourth layers

The detailed workflow of HMLET (End)

hmlet's People

Stargazers

Watchers

Forkers

hmlet's Issues

Recommend Projects

Recommend Topics

Recommend Org